JP3976920B2 - Image encoding device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an apparatus for encoding color image data.
[0002]
[Prior art]
In the conventional color image data encoding method, there is one in which color image data is decomposed into luminance information and color difference information, and each is encoded. In this encoding method, high coding efficiency can be achieved by increasing the color difference information ratio in areas where the color changes sharply, and increasing the luminance information ratio in areas where the brightness changes sharply. It was. Further, when the amount of encoded data is the same for each predetermined image unit, data access is easier, and there are advantages such as being convenient when a code is stored in a memory having a fixed capacity.
[0003]
As such conventional examples, there are techniques described in the patent publications of Patent No. 2608284 and Patent No. 2608285. In the technique described in Japanese Patent No. 2608284, first, an image is divided into blocks, and luminance is converted into information representing average, variance, and gradient of each pixel, and color differences are converted into average values. A block whose luminance variance value exceeds a predetermined value is used as a luminance edge region, and each pixel gradient information of luminance is increased, and the average value information of color difference is reduced accordingly. On the other hand, a block whose luminance variance value is lower than a predetermined value is a luminance non-edge region, and each pixel gradient information of luminance is reduced and average value information of color difference is increased accordingly.
[0004]
In the technique described in Japanese Patent No. 2608285, substantially the same conversion is performed for the luminance color difference. However, when the color difference variation is large in the block, each pixel is classified into two according to the color difference value, and the respective color difference average values are obtained. In such a block having a large variation in color difference, the average value information of the color difference is increased to two as the color difference edge region, and the average value of the luminance and the variance value information are decreased accordingly. On the other hand, a block having a small variation in color difference is a color difference non-edge region, and the average value information of color difference is reduced to one, and the average value of luminance and dispersion value information are increased accordingly. In both technologies, the bit distribution is changed according to the area, but the number of bits of the luminance color difference total is fixed, so that the encoding method is highly efficient and easy to access data.
[0005]
[Problems to be solved by the invention]
In the above-mentioned Japanese Patent No. 2608284, the luminance is separated into two regions (edge region and non-edge region), and in Japanese Patent No. 2608285, the color difference is separated into two regions (edge region and non-edge region). However, in an actual color image, two areas of luminance and two areas of color difference appear completely independently. That is, there are four areas for luminance and color difference (hereinafter, as shown in FIG. 3, these four areas are designated as [1, 0, 0] area, [1, 0, 1] area, [1, 1, 0]. ] Area, [1, 1, 1] area). Applying the above-described conventional method as it is to these four regions does not always produce good results.
[0006]
For example, the [1, 0, 0] area and the [1, 0, 1] area are the same non-edge area in terms of luminance, but the color difference differs between the non-edge area and the edge area. When the above-mentioned Japanese Patent No. 2608285 is applied, the [1, 0, 1] area decreases the bit distribution of luminance and increases the bit distribution of color difference. This is preferable for the color difference, but not preferable for the luminance, and causes image quality deterioration due to insufficient gradation. In other words, what should be changed for the two color difference areas is not the luminance / color difference bit allocation, but the color difference conversion method.
It is desirable to change the conversion method, such as creating two average values of color differences instead of one, even if the luminance and color difference bit allocation is the same for non-edge areas in the edge areas of color differences.
[0007]
In Japanese Patent No. 2608284, luminance data is converted into average, variance, and pixel gradient, which is also inferior in encoding efficiency as compared with the conversion method of the present invention described later.
SUMMARY OF THE INVENTION An object of the present invention is to provide an image coding apparatus that has high coding efficiency and easy access to code data for color image data.
[0010]
[Means for Solving the Problems]
The invention according to claim 1 is an image encoding device for processing color image data composed of luminance data and color difference data in units of blocks, and for each block, the color image data is converted into a low-frequency coefficient and a vertical frequency. Frequency coefficient conversion means for converting into four types of frequency coefficients, a high frequency coefficient in the direction, a high frequency coefficient in the horizontal direction, and a high frequency coefficient in the oblique direction, and whether each block is a chromatic color area or an achromatic color area And determining whether or not it is a luminance edge region and whether or not it is a chrominance edge region, and a region separation unit that obtains a region separation result, and the region separation unit determines that the region is a chromatic color region. For blocks, depending on whether it is a luminance edge region or a color difference edge region, the low frequency coefficient, the vertical high frequency coefficient, the horizontal high frequency coefficient, the diagonal high frequency coefficient The number of bits allocated to each of the blocks is changed, the frequency coefficient is quantized, and for the block determined to be an achromatic region, the luminance data of the block is quantized, and even if any region is determined, The object is achieved by including quantization means for obtaining a quantized output having the same number of bits for each block and encoding means for encoding the quantized output.
[0011]
According to a second aspect of the present invention, there is provided an image encoding device for processing color image data composed of luminance data and color difference data in units of blocks, and for each block, the color image data is converted into a low-frequency coefficient and a vertical length. Frequency coefficient conversion means for converting into four types of frequency coefficients, a high frequency coefficient in the direction, a high frequency coefficient in the horizontal direction, and a high frequency coefficient in the oblique direction, and whether each block is a chromatic color area or an achromatic color area And determining whether or not it is a luminance edge region and whether or not it is a chrominance edge region, and a region separation unit that obtains a region separation result, and the region separation unit determines that the region is a chromatic color region. In a block that is determined not to be a luminance edge region and not a color difference edge region, the luminance edge region is obtained by quantizing only the low frequency coefficient to obtain a quantized output. If the block is not a chrominance edge area and is not a chrominance edge area, the low frequency coefficient and the high frequency coefficient in the vertical direction depend on whether it is a luminance edge area and a chrominance edge area. For the block determined to be an achromatic region, the number of bits allocated to each of the horizontal high-frequency coefficient and the diagonal high-frequency coefficient is changed, and the frequency coefficient is quantized to obtain a quantized output. Quantize the luminance data to obtain the quantized output, (chromatic color, no luminance edge, no chrominance edge), (chromatic color, no luminance edge, with chrominance edge), (chromatic color, with luminance edge, chrominance Quantization hand that obtains quantized output with the same number of bits for each block in any area (no edge), (chromatic color, with luminance edge, with chrominance edge) and (achromatic) When, by having the encoding means for encoding the quantized output, to achieve the above object.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, preferred embodiments of the present invention will be described in detail with reference to FIGS.
FIG. 1 is a block diagram of an encoding unit (transmission side) according to the present embodiment. In this figure, RGB color image data is first converted to YCbCr color data. The conversion formula is shown below.
[0013]
[Expression 1]
Y = 0.299R + 0.587G + 0.114B
Cr = 1 / (2 (1−0.299)) (R−Y)
Cb = 1 / (2 (1-0.114)) (BY)
[0014]
Incidentally, the inverse transformation formula is as follows.
[0015]
[Expression 2]
R = Y + 2 (1−0.299) Cr
B = Y + 2 (1-0.114) Cb
G = (Y−0.299R−0.114B) /0.587
[0016]
In the above formula, Y is information representing luminance, and Cb and Cr are information representing color differences. In general, by converting from RGB to YCbCr, the entropy of the signal can be reduced, and the encoding efficiency can be increased.
[0017]
Next, the Y, Cb, and Cr signals are sent to the subband conversion unit for each 2 × 2 pixel block. FIG. 2 shows an example of subband conversion for image data. First, a low-pass filter S (n) and a high-pass filter D (n) are first applied to the original image in the horizontal direction, and then the same processing is performed in the vertical direction to obtain a horizontal high region HL, a vertical high region LH, and a diagonal high region. Four frequency band signals of HH and low band LL are generated. The simplest conversion formula is shown below as an example of the subband conversion formula.
[0018]
[Equation 3]
LPF S (n) = | (x (2n) + x (2n + 1)) / 2 |
HPF D (n) = x (2n) −x (2n + 1)
[0019]
The low frequency coefficient is the same as the average value of 4 pixels. Each high frequency coefficient represents the degree of edge in each direction, and the larger the absolute value, the steep edge is included in the corresponding block. A visual characteristic is known that the edge in the oblique direction is less perceivable than the edge in the horizontal and vertical directions. For this reason, it can be said that the diagonal high frequency coefficient is less important in image quality than the horizontal and vertical high frequency coefficients.
Using this characteristic, in the process of quantization described later, it is possible to reduce more information with the diagonal high-frequency coefficient. Since the visual redundancy can be reduced, the subband conversion according to the present embodiment is a more effective conversion method than the conventional conversion methods (average, variance, conversion to gradient of each pixel).
[0020]
Since the high frequency coefficient represents the degree of edge, for example, a block that satisfies the condition of the following equation can be classified as an edge region, and a block that does not satisfy the condition can be classified as a non-edge region.
[0021]
[Expression 4]
Luminance edge: {| LHy |> 16 or | HLy |> 16 or | HHy |> 32}
Color difference edge: {| LH _cb |> 16 or | HL _cb |> 16 or | HH _cb |> 32 or | LH _cr |> 16 or | HL _cr |> 16 or | HH _cr |> 32}
[0022]
Since the edge, non-edge 2 areas, and color difference are separated into 2 areas in the same manner, 4 areas can be separated as a combination. If both Cb and Cr representing the color difference are 0, it is determined that the region is an achromatic color region, and if not, a chromatic color region, and two regions can be further separated.
In the example shown in FIG. 3, a total of five types of region separation are performed using these combinations. The region separation unit in FIG. 1 is a part that performs this separation. The separation result (3 bits / block) is transmitted to each quantization unit and becomes a part of a finally generated code.
[0023]
The low frequency coefficient is linearized or nonlinearized to reduce the number of bits to a predetermined number. Since the YCbCr color space is not a uniform color space, nonlinear quantization is performed for the color difference.
FIG. 4 is a diagram showing the quantization characteristics of the low frequency coefficient. One of a plurality of quantization characteristics prepared in advance is selected based on the output of the image area separation unit. Accordingly, the number of bits of the quantized output changes. For example, the low frequency coefficient of luminance is 8 bits in the [1, 0, 0] area and [1, 0, 1] area, but 4 bits in the [1, 1, 0] area and [1, 1, 1] area. It becomes. The fact that the input is 8 bits and the output is 8 bits actually means that no quantization processing is performed.
[0024]
The high frequency coefficient is vector quantized. Three-dimensional vector quantization is performed in the horizontal high region, vertical high region, and diagonal high region.
FIG. 5 is a diagram illustrating an example of vector quantization. Here, for convenience of explanation, two-dimensional vector quantization of a horizontal high band and a vertical high band is shown. The plot in FIG. 5 represents the quantization representative point. Although not shown in the figure, the number of quantization representative points in the diagonal high band direction is smaller than in other directions. Similarly, for the high frequency coefficient, the quantization characteristic is selected based on the output of the region separation unit, and the number of bits of the quantization output changes. For example, the high frequency coefficient of luminance is 6 bits in the [1, 1, 0] region and [1, 1, 1] region, but 0 bits in the [1, 0, 0] region and [1, 0, 1] region. It becomes. Here, the output 0 bits means that the generated high frequency coefficient is discarded.
[0025]
Only luminance information is required in the achromatic [0, 0, 0] region. In this embodiment, the luminance subband conversion coefficient is not used, and the luminance values themselves of the four pixels in the block are linearly quantized.
The quantized output obtained in this way and the region separation result are combined in units of blocks to form a compression code. The combined compression code length is always the same in all five regions. Since the code length per block is fixed, the address calculation is facilitated when the code of an arbitrary block is extracted from all the compression codes of the image. Further, it is possible to know the total code length of the corresponding image before compression.
According to this example, the information amount per block was 96 bits before encoding, but this was encoded to 27 bits, and a compression rate of 28% was achieved.
[0026]
In the above example, the RGB color data is converted to YCbCr, but it may of course be a YIQ space used in television or an L *, a *, b * space. In addition, although region separation was performed based on the high-frequency coefficient by subband conversion, other variations of luminance values and color difference values before subband conversion (for example, the difference between the maximum value and the minimum value in a block) Can also be done. The size of the pixel block is 2 × 2 in this example, but may be 2 × 2 or more.
[0027]
【The invention's effect】
Since the invention according to the present application performs encoding of a fixed length in units of blocks, the positional relationship of the pixels in the image can be maintained when storing in the image memory. Therefore, the amount of image memory required depends on the size of the image. It can be determined uniquely. Also, when image processing is performed on this color image, information on neighboring pixels can be easily obtained, and processing can be performed in units of blocks, so that the image processing speed can be increased. Further, according to the present invention, since optimum bit allocation can be performed based on the area characteristics of luminance and color difference, respectively, high-efficiency encoding with less visual degradation utilizing human visual characteristics is possible.
[0028]
In addition, since the horizontal, vertical, and diagonal high-frequency coefficients generated by subband conversion can be quantized with different specific gravities, more efficient coding is possible.
[0029]
Further, since the separating means determines the magnitude of the variation in the value of the data in the block and separates the corresponding block into a plurality of regions according to the determination result, high coding efficiency can be achieved for the color image data. . It is possible to separate the achromatic region in the color image, the area such as high frequency of appearance black character, it is possible to perform high-quality coding.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a configuration example of an image encoding device according to the present invention.
FIG. 2 is a diagram illustrating an example of subband conversion for an image signal.
FIG. 3 is a diagram illustrating an example of bit allocation by region.
FIG. 4 is a diagram illustrating a specific example of quantization of a low frequency coefficient.
FIG. 5 is a diagram illustrating a specific example of quantization of a high frequency coefficient.

Claims (2)

An image encoding device that processes color image data composed of luminance data and color difference data in units of blocks,
Frequency coefficient conversion means for converting the color image data into four types of frequency coefficients, a low frequency coefficient, a vertical high frequency coefficient, a horizontal high frequency coefficient, and a diagonal high frequency coefficient for each block;
For each block, it is determined whether it is a chromatic color region or an achromatic color region, whether it is a luminance edge region, and whether it is a color difference edge region, and a region separation result Region separation means to obtain,
The block determined to be a chromatic color area by the area separation unit is a low frequency coefficient, a high frequency coefficient in the vertical direction, a horizontal direction depending on whether it is a luminance edge area or a color difference edge area. Changing the number of bits assigned to each of the high-frequency coefficient and the diagonal high-frequency coefficient, and quantizing the frequency coefficient,
For a block determined to be an achromatic region, the luminance data of the block is quantized,
Quantization means for obtaining a quantized output of the same number of bits for each block, regardless of which area is determined,
Encoding means for encoding the quantized output;
An image encoding apparatus comprising: An image encoding device that processes color image data composed of luminance data and color difference data in units of blocks,
Frequency coefficient conversion means for converting the color image data into four types of frequency coefficients, a low frequency coefficient, a vertical high frequency coefficient, a horizontal high frequency coefficient, and a diagonal high frequency coefficient for each block;
For each block, it is determined whether it is a chromatic color region or an achromatic color region, whether it is a luminance edge region, and whether it is a color difference edge region, and a region separation result Region separation means to obtain,
A block determined as a chromatic color region by the region separating means,
In a block that is determined not to be a luminance edge region and not a color difference edge region, only the low frequency coefficient is quantized to obtain a quantized output,
In blocks other than those determined not to be luminance edge areas and color difference edge areas,
The number of bits allocated to each of the low frequency coefficient, the high frequency coefficient in the vertical direction, the high frequency coefficient in the horizontal direction, and the high frequency coefficient in the diagonal direction depending on whether it is a luminance edge area and a color difference edge area , Quantize the frequency coefficient to obtain a quantized output,
For a block determined to be an achromatic region, quantize the luminance data of the block to obtain a quantized output,
(Chromatic color, no luminance edge, no color difference edge), (chromatic color, no luminance edge, with color difference edge), (chromatic color, with luminance edge, no color difference edge), (chromatic color, with luminance edge, with color difference edge) ) And (achromatic) in any region, quantization means for obtaining a quantized output with the same number of bits for each block;
Encoding means for encoding the quantized output;
An image encoding apparatus comprising:

Images